人脸生成（Face Generation）¶

在该项目中，你将使用生成式对抗网络（Generative Adversarial Nets）来生成新的人脸图像。

获取数据¶

该项目将使用以下数据集：

MNIST
CelebA

由于 CelebA 数据集比较复杂，而且这是你第一次使用 GANs。我们想让你先在 MNIST 数据集上测试你的 GANs 模型，以让你更快的评估所建立模型的性能。

如果你在使用 FloydHub, 请将 data_dir 设置为 "/input" 并使用 FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

探索数据（Explore the Data）¶

MNIST¶

MNIST 是一个手写数字的图像数据集。你可以更改 show_n_images 探索此数据集。

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x28c96e64860>

CelebA¶

CelebFaces Attributes Dataset (CelebA) 是一个包含 20 多万张名人图片及相关图片说明的数据集。你将用此数据集生成人脸，不会用不到相关说明。你可以更改 show_n_images 探索此数据集。

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x28c96ed7be0>

预处理数据（Preprocess the Data）¶

由于该项目的重点是建立 GANs 模型，我们将为你预处理数据。

经过数据预处理，MNIST 和 CelebA 数据集的值在 28×28 维度图像的 [-0.5, 0.5] 范围内。CelebA 数据集中的图像裁剪了非脸部的图像部分，然后调整到 28x28 维度。

MNIST 数据集中的图像是单通道的黑白图像，CelebA 数据集中的图像是三通道的 RGB 彩色图像。

建立神经网络（Build the Neural Network）¶

你将通过部署以下函数来建立 GANs 的主要组成部分:

model_inputs
discriminator
generator
model_loss
model_opt
train

检查 TensorFlow 版本并获取 GPU 型号¶

检查你是否使用正确的 TensorFlow 版本，并获取 GPU 型号

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.0.1
Default GPU Device: /gpu:0

输入（Input）¶

部署 model_inputs 函数以创建用于神经网络的占位符 (TF Placeholders)。请创建以下占位符：

输入图像占位符: 使用 image_width，image_height 和 image_channels 设置为 rank 4。
输入 Z 占位符: 设置为 rank 2，并命名为 z_dim。
学习速率占位符: 设置为 rank 0。

返回占位符元组的形状为 (tensor of real input images, tensor of z data, learning rate)。

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    input_real = tf.placeholder(tf.float32, [None, image_width, image_height, image_channels], name='input_real')
    input_z = tf.placeholder(tf.float32, [None, z_dim], name='input_z')
    learning_rate = tf.placeholder(tf.float32)

    return input_real, input_z, learning_rate


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

辨别器（Discriminator）¶

部署 discriminator 函数创建辨别器神经网络以辨别 images。该函数应能够重复使用神经网络中的各种变量。在 tf.variable_scope 中使用 "discriminator" 的变量空间名来重复使用该函数中的变量。

该函数应返回形如 (tensor output of the discriminator, tensor logits of the discriminator) 的元组。

def discriminator(images, reuse=False, alpha=0.2):
    """
    Create the discriminator network
    :param image: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    # TODO: Implement Function
    with tf.variable_scope('discriminator', reuse=reuse):
        x1 = tf.layers.conv2d(images, 64, 5, strides=2, padding='same')
        x1 = tf.maximum(alpha * x1, x1)
        # 14x14x64 now
        
        x2 = tf.layers.conv2d(x1, 128, 5, strides=2, padding='same')
        x2 = tf.layers.batch_normalization(x2, training=True)
        x2 = tf.maximum(alpha * x2, x2)
        # 7x7x128 now
        
        x3 = tf.layers.conv2d(x2, 256, 5, strides=2, padding='same')
        x3 = tf.layers.batch_normalization(x3, training=True)
        x3 = tf.maximum(alpha * x3, x3)
        # 4x4x256 now
        
        flat = tf.reshape(x3, [-1, 4*4*256])
        logits = tf.layers.dense(flat, 1)
        out = tf.sigmoid(logits)

    return out, logits


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

生成器（Generator）¶

部署 generator 函数以使用 z 生成图像。该函数应能够重复使用神经网络中的各种变量。在 tf.variable_scope 中使用 "generator" 的变量空间名来重复使用该函数中的变量。

该函数应返回所生成的 28 x 28 x out_channel_dim 维度图像。

def generator(z, out_channel_dim, is_train=True, alpha=0.2):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    # TODO: Implement Function
    with tf.variable_scope('generator', reuse=not is_train):
        x1 = tf.layers.dense(z, 4*4*512)
        x1 = tf.reshape(x1, [-1, 4, 4, 512])
        x1 = tf.layers.batch_normalization(x1, training=is_train)
        x1 = tf.maximum(alpha * x1, x1)
        # 4x4x512 now
        
        x2 = tf.layers.conv2d_transpose(x1, 256, 4, strides=1, padding='valid')
        x2 = tf.layers.batch_normalization(x2, training=is_train)
        x2 = tf.maximum(alpha * x2, x2)
        # 7x7x256 now
        # out_height = in_height * stride + filter_height - 1
        
        x3 = tf.layers.conv2d_transpose(x2, 128, 5, strides=2, padding='same')
        x3 = tf.layers.batch_normalization(x3, training=is_train)
        x3 = tf.maximum(alpha * x3, x3)
        # 14x14x128 now
        
        logits = tf.layers.conv2d_transpose(x3, out_channel_dim, 5, strides=2, padding='same')
        # 28x28xout_channel_dim now
        out = tf.tanh(logits)
    
    return out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

损失函数（Loss）¶

部署 model_loss 函数训练并计算 GANs 的损失。该函数应返回形如 (discriminator loss, generator loss) 的元组。

使用你已实现的函数：

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    # TODO: Implement Function
    g_model = generator(input_z, out_channel_dim)
    d_model_real, d_logits_real = discriminator(input_real)
    d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)
    
    g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake,
                                                                   labels=tf.ones_like(d_model_fake)))
    d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real,
                                                                         labels=tf.ones_like(d_model_real)))
    d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake,
                                                                         labels=tf.zeros_like(d_model_fake)))
    d_loss = d_loss_real + d_loss_fake
    
    return d_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

优化（Optimization）¶

部署 model_opt 函数实现对 GANs 的优化。使用 tf.trainable_variables 获取可训练的所有变量。通过变量空间名 discriminator 和 generator 来过滤变量。该函数应返回形如 (discriminator training operation, generator training operation) 的元组。

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # TODO: Implement Function
    t_vars = tf.trainable_variables()
    g_vars = [var for var in t_vars if var.name.startswith('generator')]
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
    
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)
        d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
    
    return d_train_opt, g_train_opt


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

训练神经网络（Neural Network Training）¶

输出显示¶

使用该函数可以显示生成器 (Generator) 在训练过程中的当前输出，这会帮你评估 GANs 模型的训练程度。

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

训练¶

部署 train 函数以建立并训练 GANs 模型。记得使用以下你已完成的函数：

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

使用 show_generator_output 函数显示 generator 在训练过程中的输出。

注意：在每个批次 (batch) 中运行 show_generator_output 函数会显著增加训练时间与该 notebook 的体积。推荐每 100 批次输出一次 generator 的输出。

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # TODO: Build Model
    input_real, input_z, lr = model_inputs(data_shape[1], data_shape[2], data_shape[3], z_dim)
    d_loss, g_loss = model_loss(input_real, input_z, len(data_image_mode))
    d_opt, g_opt = model_opt(d_loss, g_loss, lr, beta1)
    
    sample_z = np.random.uniform(-1, 1, size=(72, z_dim))
    steps = 0
    
    session_cfg = tf.ConfigProto()
    session_cfg.gpu_options.allow_growth = True
    with tf.Session(config=session_cfg) as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            for batch_images in get_batches(batch_size):
                steps += 1 
                
                batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim))
                
                _ = sess.run(d_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
                _ = sess.run(g_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
                
                if steps % 10 == 0:
                    train_loss_d = sess.run(d_loss, feed_dict={input_real: batch_images, input_z: batch_z})
                    train_loss_g = sess.run(g_loss, feed_dict={input_z: batch_z})
                    
                    print('Epoch: {}/{}'.format(epoch_i+1, epoch_count),
                          'Discriminator Loss: {:.4f}'.format(train_loss_d),
                          'Generator Loss: {:.4f}'.format(train_loss_g))
                    
                if steps % 100 == 0:
                    show_generator_output(sess, len(batch_images), input_z, len(data_image_mode), data_image_mode)

MNIST¶

在 MNIST 上测试你的 GANs 模型。经过 2 次迭代，GANs 应该能够生成类似手写数字的图像。确保生成器 (generator) 低于辨别器 (discriminator) 的损失，或接近 0。

batch_size = 128
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 8

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch: 1/8 Discriminator Loss: 0.7712 Generator Loss: 0.7863
Epoch: 1/8 Discriminator Loss: 0.4069 Generator Loss: 1.1708
Epoch: 1/8 Discriminator Loss: 0.0711 Generator Loss: 4.5702
Epoch: 1/8 Discriminator Loss: 0.1849 Generator Loss: 2.5178
Epoch: 1/8 Discriminator Loss: 0.2148 Generator Loss: 6.4486
Epoch: 1/8 Discriminator Loss: 1.2364 Generator Loss: 8.8684
Epoch: 1/8 Discriminator Loss: 0.7998 Generator Loss: 6.4957
Epoch: 1/8 Discriminator Loss: 0.2746 Generator Loss: 2.6754
Epoch: 1/8 Discriminator Loss: 0.1945 Generator Loss: 2.5745
Epoch: 1/8 Discriminator Loss: 0.5270 Generator Loss: 2.9053

Epoch: 1/8 Discriminator Loss: 0.1698 Generator Loss: 2.5472
Epoch: 1/8 Discriminator Loss: 0.3894 Generator Loss: 3.7007
Epoch: 1/8 Discriminator Loss: 0.1735 Generator Loss: 2.9745
Epoch: 1/8 Discriminator Loss: 0.1465 Generator Loss: 3.2396
Epoch: 1/8 Discriminator Loss: 0.4751 Generator Loss: 1.3257
Epoch: 1/8 Discriminator Loss: 0.4284 Generator Loss: 1.6698
Epoch: 1/8 Discriminator Loss: 0.1914 Generator Loss: 2.5242
Epoch: 1/8 Discriminator Loss: 0.2582 Generator Loss: 2.9337
Epoch: 1/8 Discriminator Loss: 0.1468 Generator Loss: 3.1154
Epoch: 1/8 Discriminator Loss: 0.1835 Generator Loss: 2.8811

Epoch: 1/8 Discriminator Loss: 0.4644 Generator Loss: 9.2538
Epoch: 1/8 Discriminator Loss: 0.2858 Generator Loss: 1.9509
Epoch: 1/8 Discriminator Loss: 0.1079 Generator Loss: 3.1445
Epoch: 1/8 Discriminator Loss: 0.0765 Generator Loss: 4.0934
Epoch: 1/8 Discriminator Loss: 0.1533 Generator Loss: 4.0927
Epoch: 1/8 Discriminator Loss: 0.0717 Generator Loss: 3.2260
Epoch: 1/8 Discriminator Loss: 0.0928 Generator Loss: 3.0014
Epoch: 1/8 Discriminator Loss: 0.0665 Generator Loss: 3.3072
Epoch: 1/8 Discriminator Loss: 0.3286 Generator Loss: 1.9996
Epoch: 1/8 Discriminator Loss: 0.7988 Generator Loss: 0.8227

Epoch: 1/8 Discriminator Loss: 0.1500 Generator Loss: 2.6491
Epoch: 1/8 Discriminator Loss: 0.2201 Generator Loss: 1.9240
Epoch: 1/8 Discriminator Loss: 0.3488 Generator Loss: 7.9374
Epoch: 1/8 Discriminator Loss: 2.4221 Generator Loss: 0.1384
Epoch: 1/8 Discriminator Loss: 0.6282 Generator Loss: 1.1805
Epoch: 1/8 Discriminator Loss: 0.3036 Generator Loss: 3.3867
Epoch: 1/8 Discriminator Loss: 0.7367 Generator Loss: 9.7104
Epoch: 1/8 Discriminator Loss: 0.4354 Generator Loss: 1.5168
Epoch: 1/8 Discriminator Loss: 0.3734 Generator Loss: 2.0182
Epoch: 1/8 Discriminator Loss: 0.7051 Generator Loss: 0.9198

Epoch: 1/8 Discriminator Loss: 1.0699 Generator Loss: 4.9418
Epoch: 1/8 Discriminator Loss: 0.3169 Generator Loss: 1.9286
Epoch: 1/8 Discriminator Loss: 0.4663 Generator Loss: 4.3197
Epoch: 1/8 Discriminator Loss: 0.2018 Generator Loss: 2.4555
Epoch: 1/8 Discriminator Loss: 0.3863 Generator Loss: 3.6739
Epoch: 1/8 Discriminator Loss: 0.2221 Generator Loss: 3.3437
Epoch: 2/8 Discriminator Loss: 0.3249 Generator Loss: 1.7455
Epoch: 2/8 Discriminator Loss: 0.1861 Generator Loss: 2.4934
Epoch: 2/8 Discriminator Loss: 0.1809 Generator Loss: 2.8970
Epoch: 2/8 Discriminator Loss: 0.2665 Generator Loss: 1.9064

Epoch: 2/8 Discriminator Loss: 0.1815 Generator Loss: 2.3499
Epoch: 2/8 Discriminator Loss: 0.2347 Generator Loss: 2.5085
Epoch: 2/8 Discriminator Loss: 0.2169 Generator Loss: 2.5029
Epoch: 2/8 Discriminator Loss: 0.4537 Generator Loss: 1.2533
Epoch: 2/8 Discriminator Loss: 0.2197 Generator Loss: 2.5612
Epoch: 2/8 Discriminator Loss: 0.2753 Generator Loss: 1.8224
Epoch: 2/8 Discriminator Loss: 0.1406 Generator Loss: 2.4263
Epoch: 2/8 Discriminator Loss: 0.3106 Generator Loss: 3.2685
Epoch: 2/8 Discriminator Loss: 0.3721 Generator Loss: 1.6250
Epoch: 2/8 Discriminator Loss: 0.2670 Generator Loss: 2.0194

Epoch: 2/8 Discriminator Loss: 0.2011 Generator Loss: 2.5147
Epoch: 2/8 Discriminator Loss: 0.1671 Generator Loss: 2.5787
Epoch: 2/8 Discriminator Loss: 0.2415 Generator Loss: 1.9422
Epoch: 2/8 Discriminator Loss: 0.2056 Generator Loss: 1.9763
Epoch: 2/8 Discriminator Loss: 0.1599 Generator Loss: 2.3728
Epoch: 2/8 Discriminator Loss: 0.2252 Generator Loss: 2.0031
Epoch: 2/8 Discriminator Loss: 0.3848 Generator Loss: 2.3955
Epoch: 2/8 Discriminator Loss: 0.2422 Generator Loss: 2.1003
Epoch: 2/8 Discriminator Loss: 0.1250 Generator Loss: 3.0012
Epoch: 2/8 Discriminator Loss: 0.3126 Generator Loss: 1.7205

Epoch: 2/8 Discriminator Loss: 0.1636 Generator Loss: 2.5692
Epoch: 2/8 Discriminator Loss: 0.1239 Generator Loss: 2.8038
Epoch: 2/8 Discriminator Loss: 0.1490 Generator Loss: 2.3720
Epoch: 2/8 Discriminator Loss: 0.1286 Generator Loss: 2.8222
Epoch: 2/8 Discriminator Loss: 0.0975 Generator Loss: 2.9526
Epoch: 2/8 Discriminator Loss: 0.1351 Generator Loss: 2.4623
Epoch: 2/8 Discriminator Loss: 0.0736 Generator Loss: 3.6906
Epoch: 2/8 Discriminator Loss: 0.1730 Generator Loss: 2.2617
Epoch: 2/8 Discriminator Loss: 0.1252 Generator Loss: 2.6568
Epoch: 2/8 Discriminator Loss: 7.3037 Generator Loss: 12.8793

Epoch: 2/8 Discriminator Loss: 0.4189 Generator Loss: 1.6662
Epoch: 2/8 Discriminator Loss: 0.2973 Generator Loss: 2.4840
Epoch: 2/8 Discriminator Loss: 0.2822 Generator Loss: 1.7752
Epoch: 2/8 Discriminator Loss: 0.2765 Generator Loss: 1.7657
Epoch: 2/8 Discriminator Loss: 0.2280 Generator Loss: 2.0486
Epoch: 2/8 Discriminator Loss: 0.1734 Generator Loss: 2.6795
Epoch: 2/8 Discriminator Loss: 0.1960 Generator Loss: 2.2319
Epoch: 2/8 Discriminator Loss: 0.0955 Generator Loss: 3.2155
Epoch: 2/8 Discriminator Loss: 0.1202 Generator Loss: 2.6609
Epoch: 2/8 Discriminator Loss: 0.0937 Generator Loss: 3.2764

Epoch: 2/8 Discriminator Loss: 0.2549 Generator Loss: 1.8805
Epoch: 2/8 Discriminator Loss: 0.2127 Generator Loss: 2.0073
Epoch: 2/8 Discriminator Loss: 0.2184 Generator Loss: 1.9449
Epoch: 3/8 Discriminator Loss: 0.0924 Generator Loss: 3.3391
Epoch: 3/8 Discriminator Loss: 0.1000 Generator Loss: 2.8161
Epoch: 3/8 Discriminator Loss: 0.7942 Generator Loss: 1.0075
Epoch: 3/8 Discriminator Loss: 0.3738 Generator Loss: 1.7844
Epoch: 3/8 Discriminator Loss: 0.3620 Generator Loss: 1.5093
Epoch: 3/8 Discriminator Loss: 0.2496 Generator Loss: 2.1117
Epoch: 3/8 Discriminator Loss: 0.0825 Generator Loss: 3.4082

CelebA¶

在 CelebA 上运行你的 GANs 模型。在一般的GPU上运行每次迭代大约需要 20 分钟。你可以运行整个迭代，或者当 GANs 开始产生真实人脸图像时停止它。

batch_size = 128
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 16

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch: 1/16 Discriminator Loss: 0.8914 Generator Loss: 1.0087
Epoch: 1/16 Discriminator Loss: 1.3492 Generator Loss: 0.3601
Epoch: 1/16 Discriminator Loss: 0.3419 Generator Loss: 1.9170
Epoch: 1/16 Discriminator Loss: 0.2695 Generator Loss: 6.3826
Epoch: 1/16 Discriminator Loss: 3.3514 Generator Loss: 0.0412
Epoch: 1/16 Discriminator Loss: 2.1674 Generator Loss: 0.1330
Epoch: 1/16 Discriminator Loss: 0.2825 Generator Loss: 3.2726
Epoch: 1/16 Discriminator Loss: 0.1242 Generator Loss: 7.9504
Epoch: 1/16 Discriminator Loss: 0.3839 Generator Loss: 7.2135
Epoch: 1/16 Discriminator Loss: 0.8711 Generator Loss: 6.1374

Epoch: 1/16 Discriminator Loss: 0.2254 Generator Loss: 2.5678
Epoch: 1/16 Discriminator Loss: 0.1414 Generator Loss: 2.6517
Epoch: 1/16 Discriminator Loss: 0.2961 Generator Loss: 1.6763
Epoch: 1/16 Discriminator Loss: 0.0766 Generator Loss: 4.4702
Epoch: 1/16 Discriminator Loss: 0.1384 Generator Loss: 3.1032
Epoch: 1/16 Discriminator Loss: 0.0939 Generator Loss: 3.0509
Epoch: 1/16 Discriminator Loss: 0.0542 Generator Loss: 3.6010
Epoch: 1/16 Discriminator Loss: 0.1079 Generator Loss: 2.7759
Epoch: 1/16 Discriminator Loss: 0.0977 Generator Loss: 2.7881
Epoch: 1/16 Discriminator Loss: 0.0724 Generator Loss: 3.0953

Epoch: 1/16 Discriminator Loss: 0.0650 Generator Loss: 3.3636
Epoch: 1/16 Discriminator Loss: 0.3553 Generator Loss: 1.2671
Epoch: 1/16 Discriminator Loss: 0.0246 Generator Loss: 4.1412
Epoch: 1/16 Discriminator Loss: 0.0031 Generator Loss: 12.5936
Epoch: 1/16 Discriminator Loss: 0.0460 Generator Loss: 3.5237
Epoch: 1/16 Discriminator Loss: 0.0687 Generator Loss: 2.9969
Epoch: 1/16 Discriminator Loss: 0.0965 Generator Loss: 4.1021
Epoch: 1/16 Discriminator Loss: 0.0748 Generator Loss: 12.8331
Epoch: 1/16 Discriminator Loss: 0.1981 Generator Loss: 1.7703
Epoch: 1/16 Discriminator Loss: 0.0689 Generator Loss: 2.7458

Epoch: 1/16 Discriminator Loss: 4.8101 Generator Loss: 0.0089
Epoch: 1/16 Discriminator Loss: 1.3221 Generator Loss: 0.3303
Epoch: 1/16 Discriminator Loss: 0.0566 Generator Loss: 4.0126
Epoch: 1/16 Discriminator Loss: 0.0838 Generator Loss: 3.1542
Epoch: 1/16 Discriminator Loss: 0.1570 Generator Loss: 7.1712
Epoch: 1/16 Discriminator Loss: 0.0247 Generator Loss: 4.7990
Epoch: 1/16 Discriminator Loss: 0.1958 Generator Loss: 1.9567
Epoch: 1/16 Discriminator Loss: 0.0754 Generator Loss: 3.4696
Epoch: 1/16 Discriminator Loss: 0.6638 Generator Loss: 10.4774
Epoch: 1/16 Discriminator Loss: 0.9228 Generator Loss: 12.3204

Epoch: 1/16 Discriminator Loss: 0.4433 Generator Loss: 2.4471
Epoch: 1/16 Discriminator Loss: 0.3136 Generator Loss: 1.7630
Epoch: 1/16 Discriminator Loss: 0.6567 Generator Loss: 1.6726
Epoch: 1/16 Discriminator Loss: 0.4319 Generator Loss: 1.8529
Epoch: 1/16 Discriminator Loss: 0.2431 Generator Loss: 4.1304
Epoch: 1/16 Discriminator Loss: 0.7799 Generator Loss: 6.4949
Epoch: 1/16 Discriminator Loss: 1.7151 Generator Loss: 4.8183
Epoch: 1/16 Discriminator Loss: 0.3228 Generator Loss: 5.4228
Epoch: 1/16 Discriminator Loss: 0.3989 Generator Loss: 1.2085
Epoch: 1/16 Discriminator Loss: 1.3413 Generator Loss: 0.3875

Epoch: 1/16 Discriminator Loss: 1.2055 Generator Loss: 0.4180
Epoch: 1/16 Discriminator Loss: 0.2965 Generator Loss: 2.3049
Epoch: 1/16 Discriminator Loss: 0.6323 Generator Loss: 1.9712
Epoch: 1/16 Discriminator Loss: 0.2649 Generator Loss: 2.0367
Epoch: 1/16 Discriminator Loss: 0.1365 Generator Loss: 3.8986
Epoch: 1/16 Discriminator Loss: 0.3674 Generator Loss: 2.5371
Epoch: 1/16 Discriminator Loss: 0.3759 Generator Loss: 9.1377
Epoch: 1/16 Discriminator Loss: 0.4684 Generator Loss: 1.4842
Epoch: 1/16 Discriminator Loss: 0.2547 Generator Loss: 3.4271
Epoch: 1/16 Discriminator Loss: 0.4204 Generator Loss: 2.9376

Epoch: 1/16 Discriminator Loss: 0.6957 Generator Loss: 0.8672
Epoch: 1/16 Discriminator Loss: 0.0912 Generator Loss: 7.0032
Epoch: 1/16 Discriminator Loss: 1.3373 Generator Loss: 7.1259
Epoch: 1/16 Discriminator Loss: 0.3604 Generator Loss: 1.5306
Epoch: 1/16 Discriminator Loss: 0.8978 Generator Loss: 0.7437
Epoch: 1/16 Discriminator Loss: 0.0385 Generator Loss: 5.6744
Epoch: 1/16 Discriminator Loss: 0.2492 Generator Loss: 1.8388
Epoch: 1/16 Discriminator Loss: 0.6287 Generator Loss: 0.9469
Epoch: 1/16 Discriminator Loss: 0.0704 Generator Loss: 5.7562
Epoch: 1/16 Discriminator Loss: 0.0980 Generator Loss: 3.8470

Epoch: 1/16 Discriminator Loss: 0.2141 Generator Loss: 1.9976
Epoch: 1/16 Discriminator Loss: 0.1698 Generator Loss: 6.4006
Epoch: 1/16 Discriminator Loss: 0.2047 Generator Loss: 1.9623
Epoch: 1/16 Discriminator Loss: 0.4488 Generator Loss: 1.4144
Epoch: 1/16 Discriminator Loss: 0.0443 Generator Loss: 4.3201
Epoch: 1/16 Discriminator Loss: 0.1404 Generator Loss: 3.0510
Epoch: 1/16 Discriminator Loss: 0.0305 Generator Loss: 5.4976
Epoch: 1/16 Discriminator Loss: 1.4694 Generator Loss: 0.3417
Epoch: 1/16 Discriminator Loss: 0.0367 Generator Loss: 6.9154
Epoch: 1/16 Discriminator Loss: 0.1192 Generator Loss: 9.6930

Epoch: 1/16 Discriminator Loss: 1.6084 Generator Loss: 7.9498
Epoch: 1/16 Discriminator Loss: 0.2577 Generator Loss: 2.0417
Epoch: 1/16 Discriminator Loss: 0.0206 Generator Loss: 5.3449
Epoch: 1/16 Discriminator Loss: 0.0957 Generator Loss: 3.3140
Epoch: 1/16 Discriminator Loss: 1.0946 Generator Loss: 9.1465
Epoch: 1/16 Discriminator Loss: 0.5218 Generator Loss: 1.2275
Epoch: 1/16 Discriminator Loss: 0.1582 Generator Loss: 10.6461
Epoch: 1/16 Discriminator Loss: 0.0345 Generator Loss: 9.1039
Epoch: 1/16 Discriminator Loss: 1.0057 Generator Loss: 0.6100
Epoch: 1/16 Discriminator Loss: 3.0227 Generator Loss: 0.0727

Epoch: 1/16 Discriminator Loss: 0.0438 Generator Loss: 4.0983
Epoch: 1/16 Discriminator Loss: 0.0437 Generator Loss: 4.4692
Epoch: 1/16 Discriminator Loss: 1.7705 Generator Loss: 0.2427
Epoch: 1/16 Discriminator Loss: 0.3503 Generator Loss: 12.1725
Epoch: 1/16 Discriminator Loss: 0.0282 Generator Loss: 8.6978
Epoch: 1/16 Discriminator Loss: 1.4982 Generator Loss: 8.9873
Epoch: 1/16 Discriminator Loss: 0.0328 Generator Loss: 5.9098
Epoch: 1/16 Discriminator Loss: 0.0901 Generator Loss: 5.4120
Epoch: 1/16 Discriminator Loss: 0.0778 Generator Loss: 6.0714
Epoch: 1/16 Discriminator Loss: 0.8498 Generator Loss: 0.7177

提交项目¶

提交本项目前，确保运行所有 cells 后保存该文件。

保存该文件为 "dlnd_face_generation.ipynb"，并另存为 HTML 格式 "File" -> "Download as"。提交项目时请附带 "helper.py" 和 "problem_unittests.py" 文件。